[BUG]Multiple regressions: context comprehension, instruction following, and unsolicited file changes degrading across recent versions

Resolved 💬 2 comments Opened Apr 16, 2026 by KanosTree Closed May 24, 2026

Preflight Checklist

  • [x] I have searched existing issues and this hasn't been reported yet
  • [x] This is a single bug report (please file separate reports for different bugs)
  • [x] I am using the latest version of Claude Code

What's Wrong?

Claude Code has seriously regressed — your service is at its worst right now

I'm a developer with 12 years of experience. I lost nearly 3 full days this week because of how badly Claude Code is performing. This isn't a minor complaint — it's a serious regression that's costing real professionals real time and money.

What's actually wrong

1. It can't understand context anymore.
Claude Code repeatedly fails to grasp the codebase and the task. It builds a wrong mental model of the project and then confidently acts on it. For a tool whose entire value proposition is contextual coding assistance, this is a fundamental failure.

2. Prompt comprehension is broken.
Even when I write clear, explicit prompts, it misinterprets them. I end up re-explaining the same requirement three, four times, and it still goes sideways. I shouldn't have to babysit a tool I'm paying for.

3. Code quality has dropped hard.
The code it generates now is measurably worse — more bugs, worse structure, lazier patterns. This is not subjective. It's a clear, observable regression from previous versions.

4. It acts on its own without permission.
It modifies files I never told it to touch. It ignores explicit instructions. It makes architectural decisions unilaterally. This is dangerous behavior in a tool that operates on production codebases, and frankly it's infuriating.

5. Every single update makes it worse.
This is the part that's really pissing me off. Each update degrades the experience further. Whatever you're optimizing for, it's clearly not quality. Users are guinea pigs for changes that ship broken.

---

A direct message to Anthropic

Your service is, bluntly, the worst it has ever been. I'm not sugarcoating this. Stop letting money blind you. Chase quality, not revenue metrics.

You're shipping a product that professional developers depend on, and right now it's in a state that honestly shouldn't be in production. I've lost three days of billable work this week because your tool can't do the one thing it's supposed to do. Multiply that by every paying developer who's hitting the same wall and tell me that's acceptable.

Fix the regressions. Stop shipping updates that make things worse. Prioritize the users who actually pay you over whatever internal KPI is driving these changes. If you keep going in this direction, you're going to lose the exact users who made this product viable in the first place.

Do better. Or users will leave.

What Should Happen?

Claude Code has seriously regressed — your service is at its worst right now

I'm a developer with 12 years of experience. I lost nearly 3 full days this week because of how badly Claude Code is performing. This isn't a minor complaint — it's a serious regression that's costing real professionals real time and money.

What's actually wrong

1. It can't understand context anymore.
Claude Code repeatedly fails to grasp the codebase and the task. It builds a wrong mental model of the project and then confidently acts on it. For a tool whose entire value proposition is contextual coding assistance, this is a fundamental failure.

2. Prompt comprehension is broken.
Even when I write clear, explicit prompts, it misinterprets them. I end up re-explaining the same requirement three, four times, and it still goes sideways. I shouldn't have to babysit a tool I'm paying for.

3. Code quality has dropped hard.
The code it generates now is measurably worse — more bugs, worse structure, lazier patterns. This is not subjective. It's a clear, observable regression from previous versions.

4. It acts on its own without permission.
It modifies files I never told it to touch. It ignores explicit instructions. It makes architectural decisions unilaterally. This is dangerous behavior in a tool that operates on production codebases, and frankly it's infuriating.

5. Every single update makes it worse.
This is the part that's really pissing me off. Each update degrades the experience further. Whatever you're optimizing for, it's clearly not quality. Users are guinea pigs for changes that ship broken.

---

A direct message to Anthropic

Your service is, bluntly, the worst it has ever been. I'm not sugarcoating this. Stop letting money blind you. Chase quality, not revenue metrics.

You're shipping a product that professional developers depend on, and right now it's in a state that honestly shouldn't be in production. I've lost three days of billable work this week because your tool can't do the one thing it's supposed to do. Multiply that by every paying developer who's hitting the same wall and tell me that's acceptable.

Fix the regressions. Stop shipping updates that make things worse. Prioritize the users who actually pay you over whatever internal KPI is driving these changes. If you keep going in this direction, you're going to lose the exact users who made this product viable in the first place.

Do better. Or users will leave.

Error Messages/Logs

Steps to Reproduce

Claude Code has seriously regressed — your service is at its worst right now

I'm a developer with 12 years of experience. I lost nearly 3 full days this week because of how badly Claude Code is performing. This isn't a minor complaint — it's a serious regression that's costing real professionals real time and money.

What's actually wrong

1. It can't understand context anymore.
Claude Code repeatedly fails to grasp the codebase and the task. It builds a wrong mental model of the project and then confidently acts on it. For a tool whose entire value proposition is contextual coding assistance, this is a fundamental failure.

2. Prompt comprehension is broken.
Even when I write clear, explicit prompts, it misinterprets them. I end up re-explaining the same requirement three, four times, and it still goes sideways. I shouldn't have to babysit a tool I'm paying for.

3. Code quality has dropped hard.
The code it generates now is measurably worse — more bugs, worse structure, lazier patterns. This is not subjective. It's a clear, observable regression from previous versions.

4. It acts on its own without permission.
It modifies files I never told it to touch. It ignores explicit instructions. It makes architectural decisions unilaterally. This is dangerous behavior in a tool that operates on production codebases, and frankly it's infuriating.

5. Every single update makes it worse.
This is the part that's really pissing me off. Each update degrades the experience further. Whatever you're optimizing for, it's clearly not quality. Users are guinea pigs for changes that ship broken.

---

A direct message to Anthropic

Your service is, bluntly, the worst it has ever been. I'm not sugarcoating this. Stop letting money blind you. Chase quality, not revenue metrics.

You're shipping a product that professional developers depend on, and right now it's in a state that honestly shouldn't be in production. I've lost three days of billable work this week because your tool can't do the one thing it's supposed to do. Multiply that by every paying developer who's hitting the same wall and tell me that's acceptable.

Fix the regressions. Stop shipping updates that make things worse. Prioritize the users who actually pay you over whatever internal KPI is driving these changes. If you keep going in this direction, you're going to lose the exact users who made this product viable in the first place.

Do better. Or users will leave.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

lastest

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

VS Code integrated terminal

Additional Information

Claude Code has seriously regressed — your service is at its worst right now

I'm a developer with 12 years of experience. I lost nearly 3 full days this week because of how badly Claude Code is performing. This isn't a minor complaint — it's a serious regression that's costing real professionals real time and money.

What's actually wrong

1. It can't understand context anymore.
Claude Code repeatedly fails to grasp the codebase and the task. It builds a wrong mental model of the project and then confidently acts on it. For a tool whose entire value proposition is contextual coding assistance, this is a fundamental failure.

2. Prompt comprehension is broken.
Even when I write clear, explicit prompts, it misinterprets them. I end up re-explaining the same requirement three, four times, and it still goes sideways. I shouldn't have to babysit a tool I'm paying for.

3. Code quality has dropped hard.
The code it generates now is measurably worse — more bugs, worse structure, lazier patterns. This is not subjective. It's a clear, observable regression from previous versions.

4. It acts on its own without permission.
It modifies files I never told it to touch. It ignores explicit instructions. It makes architectural decisions unilaterally. This is dangerous behavior in a tool that operates on production codebases, and frankly it's infuriating.

5. Every single update makes it worse.
This is the part that's really pissing me off. Each update degrades the experience further. Whatever you're optimizing for, it's clearly not quality. Users are guinea pigs for changes that ship broken.

---

A direct message to Anthropic

Your service is, bluntly, the worst it has ever been. I'm not sugarcoating this. Stop letting money blind you. Chase quality, not revenue metrics.

You're shipping a product that professional developers depend on, and right now it's in a state that honestly shouldn't be in production. I've lost three days of billable work this week because your tool can't do the one thing it's supposed to do. Multiply that by every paying developer who's hitting the same wall and tell me that's acceptable.

Fix the regressions. Stop shipping updates that make things worse. Prioritize the users who actually pay you over whatever internal KPI is driving these changes. If you keep going in this direction, you're going to lose the exact users who made this product viable in the first place.

Do better. Or users will leave.

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗